POPROX for Researchers

What POPROX does

POPROX supports researchers in conducting live user studies (aka A/B tests or randomized control experiments) in the area of news recommendation. News is delivered through personalized daily newsletters where experimenters control content and display.

Advantages of using POPROX

• No need to recruit users.
• No need to scrape content
• No need to manage user accounts / consent / opt-out

Things to know

• You must supply your own compute.
• You must write your own code.
• You must get human subjects approval from your own institution.

Priorities

For right now, we are prioritizing experiments from US researchers. We hope to make the service available to a global audience of researchers in the future

Timeline

• Complete intake form Intake Form to describe experiment.
  • Expect to hear back from a POPROX team member within 1 week
• Create experiment design and submit it for approval
  • Approvals may take 1-2 weeks; longer if iteration needed
  • We will also discuss feasible scheduling for your experiment
• Create the code for your experiment and experiment manifest (formal specification)
• Get IRB approval from your institution
• Test code with POPROX platform
• Run experiment / deliver newsletters
• Get results, analyze
• Publish results including datasets, manifest, and optionally code

Becoming an researcher

The first step to becoming a POPROX researcher is to fill out the Intake Form. If you'd like to consult with us before doing that, please email the team at poprox@umn.edu .

If you'd like to learn more about using the platform, see the researcher documentation

We also have the slides available from our Tutorial at RecSys 2024: slides

POPROX code

POPROX is open source software hosted on GitHub. The part of most interest to researchers will be the poprox-recommender repository, which shows how our recommendation pipeline is built. Researchers are encouraged to fork this repository to start their building their POPROX-compatible experiments.

Coming soon

Accessing data

Check back for more information about using data from POPROX to bootstrap your POPROX research.

Overview of existing experiments and results

Check back for results from POPROX experiments